AI-Washing: When Saying You Use AI Becomes A Securities Problem

Somewhere in most companies' marketing materials right now is a sentence describing a product or process as AI-powered, AI-driven, or intelligent. In the overwhelming majority of cases, nobody in the organization has asked what specific technology that sentence is asserting exists, or whether anyone could demonstrate it on request. For public companies, and increasingly for private ones raising capital, that has quietly become a compliance question rather than a copywriting one.

What AI-Washing Actually Is

The term has no precise legal definition, and it is worth being clear about that rather than treating it as a defined offence. Current legal commentary describes it as the marketing practice whereby companies misrepresent the existence, extent, or efficacy of their AI capabilities[1]. Canadian commentary frames it in similar terms: promoting products and services as AI-powered, often with little substance behind the label, and characterizes it as an old trick with a modern veneer, the same pattern that emerges whenever a concept becomes commercially valuable enough that overstating it pays[2].

The direct analogy regulators themselves have drawn is to greenwashing, and the analogy is more than rhetorical. Legal analysis notes the SEC's framework for policing AI claims finds its doctrinal roots in earlier enforcement targeting ESG misrepresentation, particularly the 2021 In re BNY Mellon Investment Adviser action, in which the firm was penalized US$1.5 million for materially misleading statements about its ESG review process for mutual funds[3]. The enforcement machinery being applied to AI claims is not new machinery; it is the same anti-fraud and disclosure apparatus, pointed at a new category of claim.

It Is Not A New Offence

This distinction matters practically because it determines what a business needs to comply with. AI-washing is generally not a standalone cause of action; it is shorthand for conduct that may violate existing securities, advertising, consumer protection, and unfair competition laws when AI claims are false, misleading, or unsupported[4]. In the US context, the Securities Act of 1933 and the Securities Exchange Act of 1934 prohibit materially false or misleading statements in communications with investors, enforced through antifraud provisions including Rule 10b-5[5].

The Canadian equivalent operates through provincial securities legislation. Statements about AI aimed at investors, shareholders, or financial markets are governed by Canadian securities laws and stock exchange reporting rules that prohibit misrepresentations, with "misrepresentation" generally defined under provincial securities acts as an untrue statement of a material fact, or the omission of a fact necessary to prevent a statement from being misleading[2]. No new statute was required for AI claims to become actionable in Canada; the existing misrepresentation framework already covers them.

The Presto Automation Settlement

The most instructive documented enforcement action is worth examining precisely. On January 14, 2025, the SEC settled charges against Presto Automation Inc., a restaurant-technology company listed on Nasdaq until September 2024, over materially false and misleading statements about critical aspects of its flagship AI product, Presto Voice, an AI-assisted speech recognition technology used to automate drive-through order-taking at quick-service restaurants[6]. Presto had become publicly traded through a September 2022 SPAC merger, making this what commentators identified as the SEC's first AI-washing enforcement action against a public reporting company, and the first against a de-SPAC company[10].

Three details matter more than the headline. First, the charges were brought under Section 17(a)(2) of the Securities Act and Section 13(a) of the Exchange Act, and the order found Presto violated the anti-fraud provisions on a negligent rather than intentional basis[11]. Intent to deceive was not required. Second, the remedy was notably light: based on Presto's financial condition, remedial acts, and cooperation, the order required only that Presto cease and desist, imposing no compliance monitor, no disgorgement, and no civil penalty[6]. Third, and most useful for a business assessing its own exposure, this was not a company with no technology at all; it was a company whose public statements about what its AI product did, and how autonomously it did it, diverged from operational reality.

The negligence standard is the point to sit with. A company does not need to have set out to mislead anyone. It needs only to have made statements about its AI that were materially inaccurate and that reasonable care would have caught, which is precisely the position a business reaches through the gradual marketing drift described in the worked case below.

The Actions That Came Before It

Presto was the first action against a public reporting company, but not the first AI-washing enforcement. SEC activity in this area began in March 2024 with simultaneous settled actions against two investment advisory firms, Delphia (USA) Inc. and Global Predictions Inc., both charged with false and misleading statements about their use of AI in investment processes[12]. Analysis of those complaints notes they detailed patterns of exaggeration rather than isolated statements, suggesting enforcement risk concentrates on systemic compliance failures rather than a single loose phrase, and both firms accepted cease-and-desist orders, censures, and civil penalties[12].

A subsequent and considerably starker case illustrates the far end of the spectrum. Albert Saniger, founder of the shopping app Nate, was charged with fraudulently raising over US$42 million by claiming the app used AI to process transactions when the company in fact relied on manual human workers to complete purchases, with fabricated automation metrics claiming rates above 90%[12]. Saniger faced both criminal charges from the DOJ and civil charges from the SEC, and the SEC alleged knowing or reckless violation rather than the negligence standard applied to Presto, while seeking disgorgement and civil penalties[11].

The contrast between the Presto and Saniger outcomes is genuinely instructive: the same broad category of conduct, misstating AI capability, produced a no-penalty cease-and-desist in one case and parallel criminal proceedings in the other, and the distinguishing factors were intent, scale, and cooperation rather than the subject matter itself.

CETU: A Dedicated Enforcement Unit

Institutional signals matter as much as individual cases in predicting enforcement direction. The SEC created the Cyber and Emerging Technologies Unit (CETU), whose mission expressly includes investigating AI fraud, AI-themed scams, cybersecurity deception, and false or misleading statements about emerging technologies[7]. Canadian legal commentary characterizes CETU's creation as a decisive shift, describing a transition from what began as marketing optimism into a regulated space with real enforcement teeth, and summarizing the operative principle succinctly: if you claim AI capability, you must prove it[7].

Supporting signals have accumulated. SEC examination priorities issued in late 2025 instruct examination teams to test whether operational controls actually match AI claims[8]. Enforcement has opened parallel probes coordinated with the FTC and DOJ[8], and separate analysis confirms all three agencies focusing enforcement on misleading statements involving corporate AI use[1]. Commentary from the SEC's Division of Investment Management in February 2026 stressed that firms must be able to explain model governance and human oversight rather than relying on buzzwords[8].

The CSA Position For Canadian Issuers

Canadian reporting issuers face a directly stated regulatory expectation rather than merely an inference from US enforcement. On December 5, 2024, the Canadian Securities Administrators published CSA Staff Notice and Consultation 11-348, Applicability of Canadian Securities Laws and the Use of Artificial Intelligence Systems in Capital Markets[13]. The notice dedicates an entire section to non-investment fund reporting issuers, and focuses specifically on how AI issues should be addressed in the MD&A and Annual Information Form[9].

The notice contains the single most useful sentence available to any business trying to understand what substantiation actually means in practice. Using its own worked example: if an issuer claims it uses AI systems extensively in one of its service offerings, CSA staff would expect the issuer to define what it means by "AI system," disclose how it is using AI systems, and be able to fairly and accurately substantiate the claim that it does so extensively[14]. Three separate obligations sit in that sentence: define your terms, describe the actual use, and prove the intensity adverb. Most marketing copy fails all three.

The notice further directs that disclosure be factual and balanced in order to avoid false or misleading statements about AI use or purported use, and warns specifically against overly promotional or vague statements, requiring that disclosures highlight both benefits and risks[15]. Critically, the CSA stated it will monitor issuers' continuous disclosure filings in relation to AI system use as part of its ongoing Continuous Disclosure Review Program[15], meaning this is active surveillance rather than guidance awaiting a complaint. The CSA also acknowledged the guidance reflects AI in its current state and that its views may evolve as the technology does[14].

The guidance goes further than prohibiting overstatement. It contemplates that issuers should consider, and disclose where material, the potential consequences of AI-related risks, the adequacy of preventative measures, and prior material incidents where AI system use has raised regulatory, ethical, or legal concerns along with those incidents' effects[9]. This is a meaningfully broader obligation than simply not exaggerating: it contemplates affirmative disclosure of AI-related incidents and risk-mitigation adequacy, which requires a business to actually be tracking those things internally in order to disclose them accurately.

Why AI Claims Are Likely Material

A business might reasonably ask whether an AI claim in marketing copy rises to the level of a material fact triggering securities obligations at all. Canadian legal commentary addresses this directly and unfavourably for anyone hoping the answer is no. References to AI often signal innovation, efficiency, or enhanced value and can significantly influence investor perceptions and decision-making, which makes AI-related statements highly likely to be considered material[2]. Regulators have emphasized that AI-related claims must be treated with the same rigour as other material information[2].

The practical implication is that the informal boundary many businesses maintain between "marketing language" and "disclosure" does not hold here. If a claim about AI capability is capable of moving investor perception, and regulators have said it generally is, then it is subject to the same accuracy standard as any other material statement, regardless of whether it appeared in a prospectus or on a website homepage.

What The Comment Letters Are Asking For

The most concrete guidance available on what substantiation actually looks like comes from the questions regulators are already asking. SEC staff issued AI-related comment letters through 2025 requesting more detail on development, validation, third-party dependencies, and the real operational role of AI and machine learning in the business[6].

Each of those four categories is worth reading as a preparation checklist. Development asks what was actually built and by whom. Validation asks how you know it works, which is a harder question than most companies have documented answers to. Third-party dependencies asks whether the "AI" is your technology or a licensed API from a vendor, a distinction many companies' marketing language obscures and which materially affects the accuracy of possessive phrasing like "our AI." Real operational role asks how much of the claimed function the AI actually performs versus how much is handled by conventional software, rules, or human staff, which is precisely the gap the Presto matter concerned.

The Private Litigation Tail

Regulatory enforcement is not the only, or necessarily the largest, exposure. Legal analysis notes that private claims add substantially to the risk: when a public company overstates AI capabilities in filings, earnings calls, or other public statements and the truth later emerges, investors who relied on those statements may bring securities fraud claims, sometimes as class actions, and these suits can proceed alongside or after regulatory action and be costly regardless of outcome[4]. The same commentary notes that a single misleading claim can trigger multiple categories of exposure simultaneously, meaning a company should consider how an AI statement could be viewed by regulators, investors, customers, and competitors at once[4].

Canadian readers have a nearby reference point for how this pattern develops. The greenwashing analogy has already produced Canadian class action litigation, with the Supreme Court of Canada in 2025 dismissing an application for leave to appeal in Cohen v Dollarama, limiting that particular greenwashing class action to Quebec[7]. The procedural details differ, but the trajectory, from regulatory guidance, to enforcement, to private class litigation, is the pattern AI claims appear to be following.

This Reaches Beyond Public Companies

Most coverage of AI-washing addresses reporting issuers, which leaves the majority of Canadian businesses assuming the topic does not concern them. Three considerations argue otherwise.

First, consumer protection and advertising law applies regardless of listing status. Canada's Competition Act prohibits materially false or misleading representations to the public, and the Competition Bureau has been actively examining AI's market implications, publishing a report on algorithmic pricing in late January 2026 as part of a broader examination of AI and market dynamics[2]. A private company advertising AI capability it does not have faces the same misleading-representation exposure it would face for any other unsupported product claim.

Second, private capital raising engages securities law. A private company making AI claims to prospective investors in a financing round is making statements to investors, and the misrepresentation framework does not exempt private placements.

Third, and most immediately practical for most businesses: acquirers and lenders now ask. A company whose marketing describes proprietary AI, and whose diligence reveals a licensed third-party API with a thin configuration layer, has created a credibility problem at exactly the moment credibility is most expensive, which is the subject of the worked case below.

A Worked Case: The Word That Cost A Diligence Cycle

A Canadian software company preparing for a growth financing round had described its core offering across its website, pitch materials, and a trade publication interview as using proprietary AI to predict customer churn. The underlying reality, established during technical diligence, was a logistic regression model built several years earlier by a former employee, retrained irregularly, running against a modest feature set, alongside a set of manually-maintained business rules that handled most of the actual flagging.

None of this was fabricated, and the model genuinely worked reasonably well for its purpose. The problem was the gap between three specific words and the underlying fact pattern: "proprietary" was defensible, "AI" was arguable, and "predict" oversold what a periodically-retrained regression on a thin feature set was actually doing. The diligence process consumed an additional five weeks reconciling the company's public description against its technical reality, and the resulting disclosure schedule included qualifications the founders found genuinely uncomfortable to negotiate.

The instructive part is that nobody in the company had ever decided to overstate anything. The language had accumulated over three years of marketing iterations, each individually a modest embellishment on the last, with no point at which anyone compared the current claim against the actual system. This is the ordinary mechanism by which AI-washing occurs in businesses with no intent to deceive, and it is why the substantiation exercise below is worth doing deliberately rather than assuming your description is accurate because nobody chose to make it inaccurate.

Building A Substantiation File

The defensive posture regulators are converging on is evidence-based AI claims[7], and that converts into a specific, achievable internal exercise regardless of company size.

Inventory every external AI claim. Website, marketing materials, investor decks, press releases, executive interviews, job postings, and product documentation. Most companies are surprised by the volume and by claims nobody currently in the business remembers approving.

Map each claim to a specific system. For every claim, identify the actual technology it refers to. Claims that cannot be mapped to an identifiable system are the immediate priority.

Classify build versus licensed. Determine whether each system is genuinely developed in-house or is a licensed third-party model or API, and check whether the claim's phrasing, particularly possessives like "our AI" or "proprietary," is accurate against that finding. This is one of the four things SEC comment letters explicitly ask about.

Document validation. Record how you know each system performs as claimed, what testing was done, when, and by whom. This is the category most companies find they cannot answer, and it is also the category that determines whether a claim is substantiated or merely believed.

Set a review cadence tied to marketing changes. The worked case above happened through gradual drift, not a single decision. A substantiation file reviewed only once is a snapshot that begins going stale immediately; tying review to material marketing revisions catches drift at the point it occurs.

The Limits Of This Analysis

Several caveats matter. This article describes general regulatory direction and published guidance; it is not legal advice, and the application of misrepresentation standards to any specific claim depends on facts, materiality, and jurisdiction this article cannot assess. AI-washing enforcement is developing rapidly, and the specific enforcement posture described here reflects publicly reported positions as of mid-2026, with the SEC's own agenda and Canadian regulatory developments both in motion; the CSA published Consultation Paper 51-406 on modernizing public company regulation in July 2026[7], and outcomes from that process may alter the Canadian disclosure landscape. Finally, this article draws substantially on law firm commentary and trade analysis rather than primary regulatory text in several places, and a business with a live disclosure question should work from the primary CSA notices and applicable provincial securities legislation with qualified counsel rather than from a summary.

Related Reading

AI & Automation In Finance Algorithmic Boardrooms: The Legal And Financial Reality Of AI Agents Holding Fiduciary Duties The board-level oversight duty that AI disclosure accuracy ultimately reports into. Compliance Competition Act Greenwashing And Drip Pricing Rules The Canadian substantiation regime AI-washing enforcement is directly modelled on. AI & Automation In Finance The 95% Problem: Why Most Finance AI Pilots Fail, And What The 5% Do Differently Why the gap between claimed AI capability and operational reality is so common.

Frequently Asked Questions

What is AI-washing?

Overstating, exaggerating, or inventing the existence or capability of AI in a product or business. It has no precise legal definition and is not a standalone offence; it is shorthand for conduct that may violate existing securities, advertising, and consumer protection laws when AI claims are false, misleading, or unsupported.

Has anyone actually been penalized for this?

Yes. The SEC settled charges against Presto Automation Inc. in early 2025 over materially false and misleading statements about its flagship AI product. The SEC has also created a dedicated Cyber and Emerging Technologies Unit whose mandate expressly includes false statements about emerging technologies.

Does this apply to Canadian companies?

Yes. The CSA has published guidance expecting AI disclosure to be tailored rather than boilerplate, factual, balanced, and free of exaggerated claims, and expressly identifies AI-washing as the practice to avoid. Canadian provincial securities acts already prohibit misrepresentations, so no new statute was needed for AI claims to be actionable.

We're a private company. Does this concern us?

Potentially, through three channels: Competition Act prohibitions on materially false or misleading representations apply regardless of listing status; private capital raising engages securities misrepresentation rules; and acquirers and lenders increasingly test AI claims during diligence, where a gap between marketing and technical reality is expensive at exactly the wrong moment.

What are regulators actually asking companies to substantiate?

SEC comment letters through 2025 requested detail on four things: development, validation, third-party dependencies, and the real operational role of the AI or machine learning in the business. Those four categories work well as an internal preparation checklist.

Is calling a licensed third-party API "our AI" a problem?

It can be, which is why third-party dependencies is one of the four things regulators explicitly ask about. Possessive or proprietary framing that implies in-house development of a licensed vendor model is precisely the kind of gap between claim and reality that substantiation review is designed to surface.

References

1. Global Investigations Review. (2026). US Enforcement Agencies Intensify Scrutiny of AI Washing, Americas Investigations Review 2026. globalinvestigationsreview.com/.../ai-washing
2. BCF Avocats d'affaires. (2026, April 10). Artificial Intelligence or Artificial Claims: AI Washing, The Risk Your Company May Face. bcf.ca/en/thought-leadership/.../ai-washing
3. New York State Bar Association. (2026, March 12). Regulating AI Deception in Financial Markets, citing In re BNY Mellon Investment Adviser (2021). nysba.org/regulating-ai-deception-in-financial-markets
4. Daeryun Law. (2026, June 17). AI Washing: SEC, FTC, and Litigation Risks for Companies. daeryunlaw.com/us/practices/detail/ai-washing
5. The Regulatory Review. (2026, March 7). Regulating AI Washing. University of Pennsylvania. theregreview.org/2026/03/07/seminar-regulating-ai-washing
6. The D&O Diary. (2026, February 5). Guest Post: AI, the SEC, and the 2026 Reporting Season. dandodiary.com/.../ai-the-sec-and-the-2026-reporting-season
7. McMillan LLP. (2026, February 25). AI Washing: The New Greenwashing, How the SEC's Emerging Technologies Unit Is Rewriting the Compliance Landscape. mcmillan.ca/insights/publications/ai-washing-the-new-greenwashing
8. AI CERTs News. (2026, May 16). AI Financial Disclosure Rules Tighten Under SEC 2026. aicerts.ai/news/ai-financial-disclosure-rules-tighten-under-sec-2026
9. Blakes. (2025, February 5). CSA Provides Guidance on AI Disclosures by Public Companies. blakes.com/insights/csa-provides-guidance-on-ai-disclosures-by-public-companies
10. The D&O Diary. (2025, January 21). SEC Files AI-Washing Enforcement Action Against Restaurant Technology Company. dandodiary.com/2025/01/.../sec-files-ai-washing-enforcement-action
11. Global Investigations Review. (2025, August 15). US Enforcement Agencies Intensify Scrutiny of AI Washing, on the negligence standard applied to Presto versus the knowing/reckless standard alleged against Saniger. globalinvestigationsreview.com/.../ai-washing
12. StoneTurn. (2025, October 30). Next-Generation Compliance: Preparing for Continued SEC AI Washing Enforcement, on the March 2024 Delphia and Global Predictions actions and the Nate/Saniger case. stoneturn.com/insight/next-generation-compliance-sec-ai-washing
13. Canadian Securities Administrators. (2024, December 5). CSA Staff Notice and Consultation 11-348, Applicability of Canadian Securities Laws and the Use of Artificial Intelligence Systems in Capital Markets. osc.ca/.../csa_20241205_11-348_artificial-intelligence-systems-capital-markets.pdf
14. Blakes / Lexology. (2025, February 5). CSA Provides Guidance on AI Disclosures by Public Companies, quoting the CSA's "extensively" substantiation example. blakes.com/insights/csa-provides-guidance-on-ai-disclosures
15. Ontario Securities Commission. CSA Staff Notice and Consultation 11-348, published text, on factual and balanced disclosure and Continuous Disclosure Review Program monitoring. osc.ca/en/securities-law/.../11-348

This article discusses published regulatory guidance and legal commentary and is provided for general informational purposes. It is not legal advice. Securities disclosure obligations are jurisdiction- and fact-specific and this area is developing quickly; confirm your position with qualified securities counsel before relying on any statement here.